tai42-storage-github

A GitHub-backed Storage provider for the TAI ecosystem. It stores content as files in a GitHub repository and serves the full tai42_contract.storage.Storage surface — the text methods (load / list / upload / delete / delete_dir) plus the binary/media methods (load_bytes / upload_bytes / stat).

The TAI ecosystem

TAI is an open-source runtime for MCP tools, agents, and workflows. A Storage backend is "where content physically lives" — a pluggable provider the runtime's ResourceManager loads and renders content over. This package is one such provider (GitHub); siblings back the same contract with S3 (tai42-storage-s3) and the local filesystem (tai42-storage-local). The ecosystem is open-ended: any package can back the same contract, so this repo is this provider's own full doc home, and the documentation site covers the platform-level story:

• Storage & resources concept: https://tai42.ai/concepts/storage-and-resources
• Build a storage provider (author guide): https://tai42.ai/guides/authors/storage-provider
• Ecosystem catalog: https://tai42.ai/reference/catalog

Install

Requires Python 3.13+. Install from PyPI into the environment that runs the server:

uv add tai42-storage-github

Or from source — clone this repo and add it as an editable dependency; the tai42-* dependencies resolve in-tree from the workspace.

git clone https://github.com/tai42ai/tai42   # next to your app checkout
cd /path/to/your/app
uv add --editable ../tai42/plugins/storage-github

Use

The backend is loaded by import side-effect: importing the tai42_storage_github package runs its @tai42_app.storage.register_storage decorator, making GithubStorage the active storage provider. Point a manifest's storage_module at the package:

# manifest.yml
storage_module: tai42_storage_github

Configuration

Configure through STORAGE_GITHUB_-prefixed environment variables:

The token is held as a SecretStr; it authenticates requests but never appears in a log line, repr, or error message.

Reads and writes

• Reads (load, load_bytes) go through the raw endpoint (raw.githubusercontent.com), which serves files of any size. The Contents API is deliberately not used for reads: it silently returns an empty body for a file over 1 MiB, so a large object would read as empty bytes. The raw endpoint is CDN-cached, so a read shortly after a write may return the previous content for up to ~5 minutes.
• Writes (upload, upload_bytes) go through the Contents API (base64) and are capped conservatively at 1 MiB. A larger payload raises ValueError before the request rather than truncating; the API's own 422 is the backstop.
• Listing uses the recursive Git Trees API (one request) and raises loudly on a truncated response rather than acting on a partial listing.
• stat infers content_type from the path suffix — GitHub stores no per-object content-type.

Development

uv venv --python 3.13
uv pip install --no-sources --group dev --editable .
uv run --no-sync pytest --cov --cov-report=term-missing
uv run --no-sync ruff check .
uv run --no-sync ruff format --check .
uv run --no-sync pyright

License

Apache-2.0. See LICENSE and NOTICE.